Metabolomic Profiles during Early Childhood and Risk of Food Allergies and Asthma in Multi-ethnic Children from a Prospective Birth Cohort

Abstract

Background:

There are increasing numbers of metabolomic studies in food allergy (FA) and asthma, which, however, are predominantly limited by cross-sectional designs, small sample size, and being conducted in European populations.

Objective:

To identify metabolites which are unique to and shared by children with FA and/or asthma in a racially diverse prospective birth cohort (the Boston Birth Cohort).

Methods:

Mass spectrometry-based untargeted metabolomic profiling was performed using venous plasma collected in early childhood (N=811). FA was diagnosed based on clinical symptoms consistent with an acute hypersensitivity reaction upon food ingestion and food specific-IgE > 0.35 kU/L. Asthma was defined based on physician diagnosis. Generalized estimating equations were applied to analyze metabolomic associations with FA and asthma, adjusting for potential confounders.

Results:

During a median follow-up of 11.8±5.2 years from birth, 78 children developed FA and 171 developed asthma. Androgenic and pregnenolone steroids were significantly associated with a lower risk of FA, especially for egg allergy. N,N,N-trimethyl-5-aminovalerate (OR=0.65, 95%CI=0.48–0.87) and 1-Oleoyl-2-arachidonoyl-sn-glycero-3-phosphoinositol (OR=0.77, 95%CI=0.66–0.90) were inversely associated with FA risk. Orotidine (OR=4.73, 95%CI=2.2–10.2) and 4-cholesten-3-one (OR=0.52, 95%CI=0.35–0.77) were the top two metabolites associated with risk of asthma, although they had no association with FA. In comparison, children with both FA and asthma exhibited an altered metabolomic profile that aligned with that of FA, including altered levels of lipids and steroids.

Conclusion:

In this U.S. multi-ethnic prospective birth cohort, unique and shared alterations in plasma metabolites during early childhood were associated with risk of developing FA, asthma, or both. These findings await further validation.

Graphical Abstract

Capsule Summary

Unique and shared alterations in plasma metabolites during early childhood are associated with the risk of developing food allergies, asthma, or both in this U.S. racially diverse prospective birth cohort.

Introduction

Prevalence of allergic diseases has increased worldwide over the last few decades. There is now a well-recognized time sequential progression toward allergy development, known as the allergic march. Typically, infants initially develop atopic dermatitis (AD), then progress to develop other atopic diseases, including food allergy (FA, affecting ~10% of children in U.S) and asthma (affecting up to 15% of children). These allergic diseases are primarily defined by their shared type-II immune mechanisms. However, not all infants with AD will follow the sequential patterns of the allergic march.¹ Additionally, children with different allergic diseases (i.e., FA and asthma) have markedly diverse clinical manifestations. This phenomenon suggests that different allergic diseases may also be shaped by additional unique pathogenic pathways. Further studies are needed to identify such unique pathogenic pathways and disease-specific biomarkers, which would potentially facilitate development of targeted interventions for different allergic phenotypes.

Untargeted metabolomics, a process which comprehensively and systematically quantifies metabolites derived from an organism, has been envisaged as one of the major “omics” tools to dissect altered pathways contributing to immune-related phenotypes^2–8 and to quantify the impact of gene-environment interactions.⁹ Metabolites and metabolic pathways are intricately associated with immune cell physiology, not only by providing energy and substrate for growth and survival, but also by instructing effector functions, differentiation, and gene expression.¹⁰ Some metabolic processes such as glycolysis, fatty acid and mitochondrial metabolism are recognized as crucial players in T cell activation and differentiation. ^{11, 12}

An increasing number of metabolomic studies have been conducted for allergic diseases with novel findings.^{3–8, 13–17} For example, previous studies have reported that diacylglycerols ⁶, bile acids ^{7, 14}, and sphingolipids¹⁴ were associated with altered risk of FA, and that metabolites related to oxidative stress¹⁵, bilirubin¹⁶, and polyunsaturated fatty acids (PUFA)¹⁷ were associated with altered risk of asthma.

However, most of these available studies had very limited sample sizes, were cross-sectional, and were predominantly conducted in European populations. Further studies are also needed to compare metabolomic differences in children with different allergic phenotypes to identify disease-specific biomarkers. A recent study by Crestani et al. reported disease-specific metabolomic signatures in FA and asthma, but key limitations of this study included a small sample size (n=125) and metabolomic profiles that were measured after disease diagnosis, which may lead to reverse causality.⁷ It is also likely that their findings may be confounded by restrictive diets, which are common among children with FA.

In the Boston Birth Cohort (BBC), we have previously demonstrated that lipidomic markers in maternal blood samples collected shortly after delivery and in cord blood can predict offspring risk of FA.^{18, 19} As an extension of previous studies, we conducted metabolomic profiling in venous plasma in early childhood using a state-of-the-art untargeted Metabolon platform, which is able to interrogate thousands of metabolites across diverse biochemical space. We aimed to investigate the relationships between metabolomic profiles during early childhood and development of FA and asthma in high-risk under-represented (59% self-reported non-Hispanic Black children) inner-city children and to identify metabolites specific to or shared by FA and asthma.

Materials and Methods

Study population

This study included 811 children from the BBC (registered in ClinicalTrial.gov NCT03228875). The BBC consists of a predominantly urban, under-represented (59% Black and 22% Hispanic) population residing in low-income communities in the U.S. The BBC was initiated in 1998 with rolling enrollment at the Boston Medical Center in Boston, MA. Details about this parent cohort has been reported elsewhere.²⁰ The BBC is enriched with preterm births (< 37 weeks of gestation). Pregnancies resulting from in vitro fertilization, multiple gestations (e.g., twins, triplets), fetal chromosomal abnormalities, or major birth defects were excluded. After providing written informed consent, data on maternal sociodemographic characteristics, lifestyle (including smoking, alcohol consumption and diet), and reproductive and medical history were obtained by research assistants based on standardized questionnaires. Maternal and newborn clinical information, including birth outcomes, was abstracted from electronic medical records (EMRs). Postnatal follow-up in the BBC has been ongoing since 2004. To minimize study participants’ burden, we aligned study visits with pediatric primary care visits. Trained research staff interviewed mothers using standardized questionnaires to gather important data on postnatal health (including self-reported food allergies, asthma, and pubertal development), infant feeding and home environment, and obtained anthropometric measurements of mothers and children. The study team worked with data warehouse specialists semi-annually to electronically abstract EMRs of study children, including types and dates of clinical visits, diagnoses (International Classification of diseases [ICD] codes), medications, lab results, immunization records, growth and development parameters from all types of clinical encounters. The study protocol received initial and annual approval from the Institutional Review Boards of Boston Medical Center and the Johns Hopkins Bloomberg School of Public Health.

Figure 1 presents the flow chart of participant enrollment. As of July 2020, 2,838 mothers-child pairs have been followed in the BBC and have completed postnatal questionnaire interviews (including a FA questionnaire). The recruitment target for this study was 1,796 children born between 1999 – 2013 for whom food-specific immunoglobulin E (fs-IgE) measurements were obtained during early childhood. About 985 children without metabolomic measurements during early childhood were removed from the study. The current study focused on 811 children with available data on metabolomic profiles during early childhood (1.6 ±1.1 years).

Figure 1.

Flow diagram of the study population included in this study.

Quantification of metabolites in venous plasma during early childhood

Untargeted metabolomic profiling on plasma samples was performed utilizing ultrahigh-performance liquid chromatography tandem mass spectroscopy at Metabolon, Inc. (Durham, North Carolina, USA), according to the protocol which has been reported previously²¹. Raw data was extracted, peak-identified, and processed using Metabolon’s hardware and software. Several recovery standards were added prior to the first step in the extraction process for quality control (QC) purposes. A pooled matrix sample, generated by taking a small volume of each experimental sample, served as a technical replicate to correct for batch effects and was added into each plate (8 replicates per plate, n=96 in total) as one of the QC steps. Instrument variability was determined by calculating the median relative standard deviation (RSD) for the internal standards that were added to each sample before injection into the mass spectrometers. Overall process variability was determined by calculating the median RSD for all endogenous metabolites (i.e., non-instrument standards) present in 100% of the client matrix samples. The median RSD was 8% for the internal standards and 10% for endogenous biochemicals. Values for instrument and process variability met Metabolon’s acceptance criteria. Experimental samples were randomized across the platform and were run with the pooled matrix samples spaced evenly among the injections. To remove batch variability, for each metabolite, the values in the experimental samples were divided by the median of those samples in each instrument batch, giving each batch and thus the metabolite a median of one, and any metabolite with missing values was imputed with the minimum value across all batches. The metabolomic data was then transformed using the natural log for statistical analyses.

The metabolomic dataset is comprised of a total of 1,579 biochemicals, 1,545 of which can be successfully detected in pooled matrix samples. Of these, 1,206 compounds are of known identity (named biochemicals) and 339 compounds are of unknown structural identity (unnamed biochemicals). This study focused on those compounds of known identity. After removing 181 known metabolites with high RSD (RSD > 25%), 145 metabolites with high missing rate (>50%), and 105 xenobiotic metabolites, a total of 775 metabolites with known identity were applied in subsequent analyses.

Definition of outcomes

Plasma fs-IgE in early childhood to the eight most common food allergens (hen’s egg white, cow’s milk, peanut, soy, shrimp, walnut, wheat and codfish) was measured using ImmunoCAP^® (Thermo-Fisher/Phadia) at Quest Diagnostics. Food sensitization (FS) was defined as having a fs-IgE ≥ 0.35 kU_A/L.^{22, 23} At each follow-up visit, mothers were interviewed about history of physician-diagnosed FA for the index child, clinical allergic symptoms (including anaphylaxis) ever experienced by the child upon ingestion of each of the eight allergenic foods, and the timing and treatment of these reactions (if any). A dietary history was also recorded at each visit, in which parents were asked if, and how often, the child consumed the allergenic foods. Based on these data, a child was defined as having FA if there was evidence of FS based on fs-IgE measurement (fs-IgE >0.35kU/L); and one of the following criteria was met: a) having convincing clinical allergic symptoms in skin, pharynx, oral cavity, lower respiratory and/or gastrointestinal tract, within 2 hours upon ingestion of that food; or b) having fs-IgE ≥ 95% positive predictive value (PPV) and avoidance of that food (thus, no report of clinical allergic symptoms due to no ingestion of that food), for a subset of children without specific food exposure ^{18, 24}. For a sensitivity analysis, we defined FA cases using more stringent criteria: meeting the FA definition above, plus fs-IgE ≥ 95% PPV to the specific food.

In this study, asthma was diagnosed at five years of age or older, based on physician diagnoses and ICD codes (ICD-9: 493.0–9; ICD-10: J45) from each child’s medical records.

Covariates

At enrollment, mothers were interviewed to collect demographic characteristics and exposures (including maternal self-reported race and ethnicity, marital status, educational level, parity, smoking during pregnancy, and maternal age at delivery) before and during pregnancy. Maternal pre-pregnancy body mass index (BMI) was calculated as self-reported pre-pregnancy weight in kilograms divided by self-reported height in meters squared. Maternal history of atopy was defined if the mother ever had any of the physician-diagnosed allergic conditions including FA, AD, atopic rhinitis, or asthma. Maternal Mediterranean diet score during pregnancy was imputed based on food frequency questionnaire, as reported previously²⁵. During the follow-up visit at child’s age < 2 years, each mother was interviewed about the following variables: 1) breastfeeding history (“bottle-fed exclusively”, “both bottle-fed and breastfed”, or “breastfed exclusively”), which was grouped into “never” versus “ever” breastfeeding, as reported previously;²⁶ and 2) timing of first solid food introduction, which was classified into three groups: within the first 3 months; 4–6 months; and later than 7 months. AD during the first year of life was further defined based on physician diagnosis. A few categorical covariates (i.e., maternal history of atopy and timing of first solid food introduction) which had a limited number of missing data were imputed with another fixed value (“missing” in this study).

Statistical analysis

Population characteristics was presented and compared in children with and without FA, and in children with and without asthma, based on t-tests and chi square tests, respectively, for continuous and categorical variables. Given that there were 13 sibling pairs enrolled, generalized estimating equation (GEE) models were applied to investigate metabolomic associations with each allergic outcome, with the adjustment of known or suspected confounders. When FA was analyzed as the outcome, we adjusted for covariates in the model including maternal race and ethnicity (non-Hispanic Black vs others), maternal history of atopy, child’s age at metabolomic profiling, child’s age at the last questionnaire interview, and other variables that were different between children with and without FA (maternal Mediterranean diet score during pregnancy, child’s sex, timing of first solid food introduction). When asthma was analyzed as the outcome, we adjusted for maternal race and ethnicity, maternal history of atopy, child’s age at metabolomic profiling, child’s age at the last clinical visit and other variables that were significantly different between children with and without asthma (maternal age at delivery, maternal smoking during pregnancy, maternal particulate matter (PM_2.5) exposure during pregnancy, gestational age at delivery, and breastfeeding). We further applied Cox proportional hazards models to generate hazard ratios (HR) with time-to-asthma as the outcome, with years of follow-up since the date of plasma collection for metabolomic profiles as the time scale. We followed children until date of asthma diagnosis, or until date of last visit, whichever came first, with the adjustment of all the above-mentioned covariates except child’s age at the last clinical visit.

There are no consensus standards for multiple testing correction in metabolomic research, and methods such as the Bonferroni correction and false discovery rate are considered too stringent for metabolomic data due to the high correlations between functionally related metabolites. We thus employed a false discovery correction using a “number of effective tests” method to account for the highly collinear nature of metabolites, which has been applied in metabolomic studies ^{27, 28}. To define “number of effective tests”, we used principal component analysis of the metabolites to determine how many components are required to explain 50% of the variance in the data. In this study, 13 components were needed to account for >50% of variance, and thus 13 was regarded as the number of effective tests and was used to compute a multiple testing corrected p value cutoff as 0.05/13=0.00385.

To explore shared as well as disease-specific metabolomic profiles, we then generated a composite outcome based on FA and asthma status: neither (the control group), FA only, asthma only, or both FA and asthma. Multinomial regression models were applied to test the metabolomic associations with this four-level categorical outcome, adjusting for maternal race and ethnicity, maternal history of atopy, child’s age at metabolomic profiling, child’s age at last clinical visit and other covariates that were significantly different among the four groups, including: gestational age at delivery, child’s sex, breastfeeding, timing of first solid food introduction.

Results

In this study, 811 children were followed from birth for an average of 11.8 (SD: 5.2) years, of whom 78 were allergic to one or more foods at about 3.3 (SD: 2.2) years of age, 704 were non-allergic (including 141 with asymptomatic FS), and 29 had some clinical symptoms or a physician diagnosis of FA (based on ICD codes) but did not meet our FA criteria. These 29 children were not included in the analyses with FA as the outcome. There were 679 children who were followed for more than 5 years with EMR data for asthma diagnosis; and 171 with a diagnosis of asthma (Figure 1). Plasma collection for metabolomic profiling was conducted at 1.6 (SD: 1.1) years of age.

Metabolomic associations with FA

Population characteristics of the enrolled children with and without FA are shown in Table 1. Compared to those without FA, children with FA were more likely to be male. Additionally, food-allergic children were more likely to be born to mothers with a lower Mediterranean diet score during pregnancy, to develop early AD during the first year and to have delayed first solid food introduction, compared to children without FA.

Table 1.

Population characteristics of children with and without food allergy, as well as of children with and without asthma

Population Characteristics a	No FA	FAb	No Asthma	Asthmac
N	704	78	508	171
Maternal age at delivery, years	28.5 ± 6.3	28.5 ± 6.8	28.9 ± 6.4	27.7 ± 6.9*
Gestational age at delivery, weeks	38.4 ± 2.6	38.6 ± 2.7	38.6 ± 2.3	37.8 ± 3.0**
Race and ethnicity, non-Hispanic Black,n(%)	407 (57.8)	53 (67.9)	306 (60.2)	111 (64.9)
Maternal marital status, married, n (%)	239 (33.9)	19 (24.4)	176 (34.6)	48 (28.1)
Maternal educational status, n (%)
< High School	206 (29.3)	19 (24.4)	148 (29.1)	51 (29.8)
High School	250 (35.5)	33 (42.3)	196 (38.6)	64 (37.4)
College or above	246 (34.9)	25 (32.0)	163 (32.1)	54 (31.6)
Missing	2 (0.3)	1 (1.3)	1 (0.2)	2 (1.2)
Maternal prepregnancy BMI categories, n(%)
Normal	324 (46.0)	38 (48.7)	230 (45.3)	73 (42.7)
Overweight	194 (27.6)	18 (23.1)	137 (27.0)	51 (29.8)
Obese	152 (21.6)	17 (21.8)	116 (22.8)	37 (21.6)
Missing	34 (4.8)	5 (6.4)	25 (4.9)	10 (5.9)
Nulliparity, n (%)	415 (58.9)	43 (55.1)	314 (61.8)	93 (54.4)
Maternal smoking during pregnancy, n (%)
Never	575 (81.7)	67 (85.9)	437 (86.0)	131 (76.6) *
Quitter	57 (8.1)	6 (7.7)	34 (6.7)	18 (10.5)
Continuous	72 (10.2)	5 (6.4)	37 (7.3)	22 (12.9)
Maternal Mediterranean diet score	24.9 ± 4.0	23.9 ± 4.5*	24.9 ± 4.2	24.4 ± 3.7
PM2.5 exposure during pregnancy, μg/m3	8.9 ±1.4	8.8 ± 1.2	8.8 ± 1.4	9.0 ±1.5*
Maternal diabetes, n (%)
None	614 (87.2)	64 (82.1)	440 (86.6)	147 (86.0)
Gestational	59 (8.4)	9 (11.5)	40 (7.9)	17 (9.9)
Pregestational	31 (4.4)	5 (6.4)	28 (5.5)	7 (4.1)
Maternal history of atopy, n (%)
No	467 (66.3)	44 (56.4)	352 (69.3)	95 (55.6)***
Yes	158 (22.5)	22 (28.2)	95 (18.7)	65 (38.0)
Missing	79 (11.2)	12 (15.4)	61 (12.0)	11 ( 6.4)
Delivery type, vaginal, n (%)	458 (65.1)	52 (66.7)	331 (65.2)	106 (62.0)
Preterm birth, n (%)	136 (19.3)	13 (16.7)	80 (15.7)	46 (26.9) **
Child’s sex, male, n (%)	342 (48.6)	51 (65.4) **	244 (48.0)	96 (56.1)
Child’s age at metabolomic profiling, years	1.7 ± 1.1	1.5 ± 1.0	1.7 ± 1.1	1.7 ± 1.2
Never breastfed, n (%)	171 (24.3)	15 (19.2)	106 (20.9)	53 (31.0) **
AD during the 1st year, n (%)	77 (10.9)	28 (35.9)***	69 (13.6)	28 (16.4)
Timing of first solid food introduction, n (%)
≤3 months	143 (20.3)	8 (10.3) **	98 (19.3)	33 (19.3)
4–6 months	477 (67.8)	56 (71.8)	352 (69.3)	110 (64.3)
≥7 months	80 (11.4)	11 (14.1)	55 (10.8)	24 (14.0)
Missing	4 (0.6)	3 (3.8)	3 (0.6)	4 (2.3)
Child’s age at last clinical visit, years	11.6 ± 5.4	13.0 ± 4.6 *	13.2 ± 3.7	14.8 ± 2.9***

^aMean±SD and n (%) are shown for continuous and categorical variables, respectively.

^bAbout 29 children who had some clinical symptoms on food exposure (or had physician diagnosed FA) but did not meet the definition criteria of FA were removed from the analyses with FA risk as the outcome.

^cAbout 132 children who had no available EMR data at 5 years or older were removed from the analyses with asthma risk as the outcome.

^{*,**, ***} The differences in population characteristics between children with and without FA, or between children with and without asthma were tested using the t-test and chi square test for continuous and categorical variables, respectively.

^*P < 0.05

^**P <0.01

^***P < 0.001

After adjustment for covariates including those that significantly differed between children with and without FA as mentioned above, we observed that 8 metabolites, including 4 androgenic steroids, two pregnenolone steroids, 1-oleoyl-2-arachidonoyl-sn-glycero-3-phosphoinositol (1-oleoyl-2-arachidonoyl-GPI) (18:1/20:4) and N,N,N-trimethyl-5-aminovalerate (TMAVA), were inversely associated with risk of FA (Table 2, and Figure 2, all P < 0.00385). These associations remained comparable with additional adjustment for maternal triacylglycerol metabolites (C48:1 and C58:10) which were significantly associated with FA¹⁸, or with additional adjustment for early AD during the first year of life (data not shown); or when those children who did not have FA but developed other allergic conditions (asthma, allergic rhinitis, and AD) were removed from the control group (data not shown). The associations for the same eight metabolites remained largely unchanged when 16 children with FA diagnosed before the time of metabolomic profiling were removed from the analyses (Table E1); or when 72 children having a history of oral or inhaled or injected steroid treatment (abstracted from EMRs) before the time of metabolomic profiling were removed from the analyses (Table E1).

Table 2.

Early childhood metabolomic associations with risk of overall food allergy in the Boston Birth Cohort

Metabolites	FA		FA, with stringent definition a		Pathways
	OR (95% CI)b	P	OR (95% CI)b	P
Dehydroepiandrosterone sulfate (DHEA-S)	0.74 (0.62–0.88)	0.0007	0.61 (0.49–0.76)	1.3×10−5 c	Androgenic Steroids
Androsterone sulfate	0.73 (0.60–0.88)	0.0008	0.56 (0.44–0.72)	6.1×10−6 c	Androgenic Steroids
Androstenediol (3alpha, 17alpha) monosulfate	0.71 (0.57–0.89)	0.0033	0.56 (0.42–0.76)	0.0002 c	Androgenic Steroids
Androstenediol (3beta,17beta) monosulfate	0.78 (0.66–0.92)	0.0035	0.64 (0.51–0.82)	0.0003c	Androgenic Steroids
1-oleoyl-2-arachidonoyl-GPI (18:1/20:4)*	0.77 (0.66–0.90)	0.0008	0.82 (0.67–1.00)	0.0556	Phosphatidylinositol (PI)
Pregnenediol sulfate *	0.61 (0.44–0.84)	0.0031	0.49 (0.32–0.76)	0.0012	Pregnenolone Steroids
Pregnenolone sulfate	0.66 (0.50–0.87)	0.0036	0.56 (0.39–0.81)	0.0021	Pregnenolone Steroids
N,N,N-trimethyl-5-aminovalerate	0.65 (0.48–0.87)	0.0033	0.58 (0.40–0.82)	0.0025	Lysine Metabolism
5alpha-pregnan-3beta, 20alpha-diol monosulfate	0.67 (0.50–0.88)	0.0048	0.50 (0.35–0.71)	0.0001c	Progestin Steroids
5alpha-pregnan-3beta, 20beta-diol monosulfate	0.69 (0.52–0.93)	0.0138	0.54 (0.36–0.79)	0.0016	Progestin Steroids
5alpha-pregnan-3beta, 20alpha-diol disulfate	0.68 (0.46–1.01)	0.0559	0.43 (0.26–0.72)	0.0013	Progestin Steroids
Androstenediol (3beta,17beta) disulfate	0.76 (0.63–0.92)	0.0051	0.65 (0.51–0.82)	0.0004c	Androgenic Steroids
Pregnenediol disulfate *	0.65 (0.45–0.93)	0.0177	0.48 (0.30–0.75)	0.0014	Pregnenolone Steroids
21-hydroxypregnenolone disulfate	0.63 (0.46–0.87)	0.0054	0.52 (0.35–0.79)	0.0021	Pregnenolone Steroids
Metabolonic lactone sulfate	0.69 (0.49–0.97)	0.0342	0.48 (0.30–0.77)	0.0022	---
Cyclo(gly-pro)	1.14 (0.87–1.50)	0.3440	1.47 (1.13–1.90)	0.0037	Dipeptide

A metabolite name with an asterisk * indicates that the identity of the metabolite has not been confirmed.

^aFA cases was defined as having clinical symptoms within 2 hours of ingestion to a specific food, plus having sensitization to the same food allergen with fs-IgE > 95% positive predictive value to the specific food.

^bAdjusted for maternal race and ethnicity (non-Hispanic Black vs others), maternal history of atopy, maternal Mediterranean diet score during pregnancy, child’s age at metabolomic profiling, child’s sex, timing of first solid food introduction, and child’s age at the last questionnaire interview.

^cFalse discovery rate (FDR) < 0.05

Figure 2.

Volcano plot for the associations between metabolomic profiles during early childhood and risk of food allergy and asthma, respectively. Figure 2A presents metabolomic associations with food allergy, adjusting for maternal race and ethnicity (non-Hispanic Black vs others), maternal history of atopy, maternal Mediterranean diet score during pregnancy, child’s age at metabolomic profiling, child’s sex, timing of first solid food introduction, and child’s age at the last questionnaire interview. Figure 2B presents metabolomic associations with asthma, adjusting for maternal age at delivery, maternal race and ethnicity, maternal smoking during pregnancy, maternal history of atopy, maternal particulate matter (PM_2.5) exposure during pregnancy, gestational age at delivery, breastfeeding, child’s age at metabolomic profiling, and child’s age at the last clinical visit.

When FA cases were limited to those cases who had fs-IgE > 95% PPV (n=45 cases), similar associations were observed for the same eight metabolites as mentioned above. Significant associations were also noticed for another three androgenic /pregnenolone steroids, three progestin steroids, metabolomic lactone sulfate and one dipeptide (Table 2). Of note, with this stringent definition of FA, five androgenic steroids and one progestin steroid remained significant after FDR correction for multiple testing, all of which were inversely associated with risk of FA (all P < 0.0004, Table 2).

Given that the steroid metabolites are highly correlated with child’s age, we performed sensitivity analyses in 353 children with metabolomic profiling conducted within the first year of life (Table E2). Similar associations with FA were observed for the 8 identified FA-associated metabolites. Additionally, within this subset, another 12 metabolites were identified as having significant associations with FA, including three progestin steroids and another androgenic steroid, which were inversely associated with risk of FA, as well as 8 others that were positively associated with risk of FA (taurocholate and taurohyocholate in bile acid metabolism; palmitoylcholine, arachidonoylcholine and arachidoylcarnitine in fatty acid metabolism; lactosyl-N-palmitoyl-sphingosine; uracil; and N-stearoylserine. Table E2).

When allergy to each specific food was analyzed as the outcome separately, we identified two metabolites associated with peanut allergy, 8 metabolites with egg allergy, and 12 metabolites with cow’s milk allergy (Table E3). Interestingly, five out of 8 metabolites associated with hen’s egg allergy were androgenic steroids (P =0.0036 ~ 0.0001), which only showed modest associations with allergy to the other two foods (cow’s milk and peanut, Table E3). In comparison, 5 out of 12 metabolites associated with cow’s milk allergy were lipids (including 1-oleoyl-2-arachidonoyl-GPI, 2 sphingomyelins, 1 lysophospholipid and 1 diacylglycerol (DAG)) and 3 were involved in secondary bile acid metabolism. N,N,N-trimethyl-5-aminovalerate (TMAVA), which was involved in lysine metabolism, had significant associations with allergies to both cow’s milk (OR=0.30, 95%CI=0.18–0.49, P=1.5×10⁻⁶) and peanut (OR=0.50, 95%CI=0.34–0.73, P=0.0004). The associations between TMAVA and risk of cow’s milk allergy remained significant after FDR correction.

Metabolomic associations with asthma

Population characteristics of the enrolled children with and without physician-diagnosed asthma is shown in Table 1. Mothers of children with asthma were younger at time of delivery, exposed to higher levels of air pollution (PM_2.5) during pregnancy, more likely to smoke during pregnancy, and more likely to have atopic phenotypes than mothers of children without asthma. Children with asthma were more likely to be born prematurely and be bottle-fed compared to children without asthma (all P < 0.05).

When physician-diagnosed asthma was analyzed as the outcome, five metabolites were identified, four of which (orotidine, N-acetylmethionine, isovalerylglycine and suberate) were positively and one of which (4-cholesten-3-one) was inversely associated with risk of asthma. Orotidine remained to be positively associated with asthma risk after FDR adjustment for multiple testing (OR=3.8, 95%CI=2.0–7.1, P=4.5×10⁻⁵). When Cox proportional Hazard regression models were applied to identify metabolomic associations with time-to-asthma events, similar results were observed for these five metabolites. In addition, we identified that sphingosine was positively, while two progestin steroids (5alpha-pregnan-3beta,20beta-diol monosulfate and 5alpha-pregnan-3beta, 20alpha-diol monosulfate) were inversely associated with risk of developing asthma (Table 3). These two progestin steroids also showed significant associations with FA with the stringent definition (Table 2).

Table 3.

Early childhood metabolomic associations with risk of asthma diagnosed at 5 years or older in the Boston Birth Cohort

Metabolites	GEE model		Cox proportional hazard regression		Pathways
	OR (95%CI) a	P	HR (95%CI) a	P
Orotidine	3.76 (1.99–7.10)	4.5×10−5b	3.17 (1.92–5.26)	7.2×10−5b	Pyrimidine Metabolism, Orotate containing
4-cholesten-3-one	0.53 (0.37–0.75)	0.0004	0.63 (0.48–0.82)	0.0006	Sterol
N-acetylmethionine	1.88 (1.26–2.80)	0.0020	1.66 (1.08–2.55)	0.0198	Methionine, Cysteine, S-adenosylmethionine, and Taurine Metabolism
Isovalerylglycine	1.39 (1.12–1.73)	0.0028	1.24 (1.04–1.48)	0.0183	Leucine, Isoleucine and Valine Metabolism
Suberate (C8-DC)	1.47 (1.14–1.88)	0.0028	1.36 (1.11–1.68)	0.0032	Fatty Acid, Dicarboxylate
Sphingosine	1.44 (1.11–1.86)	0.0058	1.38 (1.12–1.69)	0.0023	Sphingosines
5alpha-pregnan-3beta,20beta-diol monosulfate	0.73 (0.57–0.91)	0.0066	0.73 (0.61–0.88)	0.0011	Progestin Steroids
5alpha-pregnan-3beta, 20alpha-diol monosulfate	0.74 (0.59–0.92)	0.0074	0.76 (0.64–0.91)	0.0027	Progestin Steroids

^aAdjusted for maternal age at delivery, maternal race and ethnicity, maternal smoking during pregnancy, maternal history of atopy, maternal particulate matter (PM_2.5) exposure during pregnancy, gestational age at delivery, breastfeeding, child’s age at metabolomic profiling, and child’s age at the last clinical visit (for GEE model only).

^bFalse discovery rate < 0.05

Metabolomic associations with combined FA and asthma phenotypes

We then classified all children with known FA and asthma diagnosis (N=650) into four categories based on distinct allergic status: children having neither FA nor asthma (the control group), children with FA only, children with asthma only, and children with both asthma and FA, with population characteristics in each category shown in Table E4. With multinomial regression models to adjust for covariates, we found that orotidine, 4-cholesten-3-one, sphingosine and N-acetylmethionine were significantly associated with asthma only. Another three metabolites, including lyxonate, 3-carboxy-4-methyl-5-propyl-2-furanpropanoate (CMPF), and hydroxy-CMPF, were significantly associated with risk of FA only. There were 7 metabolites that were significantly and inversely associated with overlapping allergic phenotypes (both FA and asthma), including 1-oleoyl-2-arachidonoyl-GPI and two androgenic steroids (androsterone sulfate, androstenediol (3 alpha, 17 alpha) monosulfate) which were also identified as having associations with FA (Table 4); as well as two DAGs (linoleoyl-linoleoyl-glycerol and linoleoyl-arachidonoyl-glycerol), one amino acid (2-ketocaprylate), and glycine conjugate of C10H14O2.

Table 4.

Early childhood metabolomic associations with combined FA and asthma phenotypes in the Boston Birth Cohort

Metabolites	Asthma only (N=116)		FA only (N=30)		Both Asthma and FA (N=41)		Pathways
	OR (95%CI)a	P	OR (95%CI)a	P	OR (95%CI)a	P
Orotidine	4.69 (2.23–9.85)	4.5×10−5	0.66 (0.23–1.92)	0.4467	1.98 (0.68–5.81)	0.2130	Pyrimidine Metabolism, Orotate containing
4-cholesten-3-one	0.53 (0.36–0.79)	0.0018	1.36 (0.62–3.00)	0.4484	0.57 (0.31–1.05)	0.0720	Sterol
Sphingosine	1.57 (1.16–2.12)	0.0035	0.87 (0.52–1.46)	0.5954	1.40 (0.88–2.23)	0.1533	Sphingosines
N-acetylmethionine	2.58 (1.36–4.87)	0.0036	1.34 (0.55–3.27)	0.5198	1.22 (0.54–2.78)	0.6360	Methionine, Cysteine, S-adenosylmethionine, and Taurine Metabolism
Lyxonate	1.02 (0.70–1.47)	0.9362	2.41 (1.40–4.13)	0.0015	0.88 (0.50–1.56)	0.6694	Pentose Metabolism
3-carboxy-4-methyl-5-propyl-2-furanpropanoate (CMPF)	0.95 (0.80–1.14)	0.5886	0.56 (0.39–0.80)	0.0015	0.76 (0.57–1.01)	0.0542	Fatty Acid, Dicarboxylate
Hydroxy-CMPF*	0.94 (0.76–1.15)	0.5326	0.56 (0.38–0.82)	0.0026	0.69 (0.50–0.96)	0.0276	Fatty Acid, Dicarboxylate
1-oleoyl-2-arachidonoyl-GPI (18:1/20:4)*	0.91 (0.77–1.08)	0.2697	0.86 (0.65–1.12)	0.2618	0.67 (0.53–0.83)	0.0003	Phosphatidylinositol (PI)
Androsterone sulfate	0.94 (0.78–1.13)	0.4867	0.83 (0.61–1.13)	0.2337	0.63 (0.48–0.82)	0.0007	Androgenic Steroids
Androstenediol (3alpha, 17alpha) monosulfate	0.91 (0.72–1.15)	0.4260	0.83 (0.57–1.21)	0.3215	0.58 (0.42–0.82)	0.0019	Androgenic Steroids
2-ketocaprylate	1.11 (0.77–1.59)	0.5819	1.00 (0.53–1.90)	0.9890	0.38 (0.21–0.70)	0.0019	Leucine, Isoleucine and Valine Metabolism
Glycine conjugate of C10H14O2 *	1.12 (0.80–1.59)	0.5057	1.39 (0.75–2.58)	0.2913	0.48 (0.29–0.78)	0.0030	--
Linoleoyl-linoleoyl-glycerol (18:2/18:2)*	0.88 (0.72–1.09)	0.2433	1.10 (0.73–1.66)	0.6420	0.67 (0.51–0.88)	0.0037	Diacylglycerol
Linoleoyl-arachidonoyl-glycerol (18:2/20:4)*	0.95 (0.74–1.21)	0.6697	1.06 (0.68–1.66)	0.7941	0.62 (0.45–0.86)	0.0038	Diacylglycerol

A metabolite name with an asterisk* indicates that the identity of the metabolite has not been confirmed.

^aAdjusted for maternal race and ethnicity, maternal age at delivery, maternal history of atopy, child’s age at metabolomic profiling, child’s sex, gestational age at delivery, timing of first solid food introduction, and child’s age at last clinical visit.

Discussion

In this multi-ethnic birth cohort, we performed untargeted plasma metabolomic profiling to characterize the plasma metabolome during early childhood that is associated with risk of developing FA and asthma. To our knowledge, this represents the first study of this kind in a predominantly self-reported Black and Hispanic prospective birth cohort, which is in alignment with the NIH’s calls to increase representation of under-represented, under-reported, and under-studied (U3) populations. Our analyses suggest androgenic, pregnenolone, and progestin steroids as the most significant metabolites which were inversely associated with FA. The metabolomic signature for FA may vary by the specific food allergens triggering allergy, with steroid metabolites as the main metabolites associated with hen’s egg allergy, and lipids as the main metabolites associated with cow’s milk allergy. We also identified that orotidine and 4-cholesten-3-one represent the top two metabolites associated with risk of asthma, neither of which had an association with FA, while steroid metabolites and 1-oleoyl-2-arachidonoyl-GPI were associated with risk of having both FA and asthma, which is most similar to the metabolites found to be associated with FA.

In this study, the top metabolites associated with risk of FA (especially, risk of egg allergy) were endogenous steroid metabolites (including androgenic, pregnenolone, and progestin steroids), including dehydroepiandrosterone sulfate (DHEA-S) which was the biomarker for adrenal suppression.²⁹ Consistently, DHEA-S and other related steroids has been reported to be inversely associated with multiple allergic diseases^{30, 31}. Previous studies have also reported that plasma steroid metabolites may be altered by previous steroid treatment. ^{31, 32} In this study, we observed that two metabolites (pregnanediol sulfate and pregnenolone sulfate) were significantly lower in those children with a history of steroid treatment. However, the associations between steroid metabolites and risk of FA remained largely unchanged after removing those children with a history of steroid treatment before the time for metabolomic profiling. Although the mechanisms underlying the inverse relationships between these steroid metabolites and risk of FA await further investigation, animal studies have provided evidence that DHEA-S plays a role in regulating type 2 inflammation.³³ Besides, pregnenolone, a precursor for the synthesis of corticosteroid and androgens, has been used as an anti-inflammatory agent for several decades.

As found in previous publications, a hallmark of the FA metabolomic signature is the presence of distinct alterations in lipid metabolites, including lysophospholipids ⁷, sphingomyelins (SMs) ⁷, ceramides ⁷, DAG and triacylglycerols ^{18, 19}. In the current study, although only 1-oleoyl-2-arachidonoyl-GPI showed a significant association with overall FA, more lipid metabolites were identified in associations with risk of cow’s milk allergy, including one DAG (palmitoyl-linoleoyl-glycerol), one lysophospholipid (1-margaroyl-glycerophosphoethanolamine) and two SMs. These findings further suggest that lipid metabolites during early childhood may play a role in FA (especially cow’s milk allergy) pathology, and that such impacts may be dependent on the specific food allergen triggering allergy. The identified lipid – FA associations in this study are in line with recent reports ⁷ and biologically plausible. For example, SMs are known to play vital roles in cell signaling, immunity, and inflammation and have been implicated in various disorders including atopic diseases.³⁴ Mechanistically, SMs might impact FA via differential regulation of the invariant natural killer T (iNKT) cell compartment ³⁵. iNKT cells play a role in FA as suggested by reports of decreased iNKT cell number in children with cow’s milk allergy ³⁶.

Bile acids are another category of metabolites that have been linked with allergic diseases, but the available findings were inconclusive. Lee-Sarwar et al. reported that fecal bile acids at age 1 year, including 2 primary bile acids (chenodeoxycholate and cholate), 4 conjugated bile acids (e.g., glycocholate, taurocholate), and 11 secondary microbially-modified bile acids, were reduced among study participants who later developed FA¹⁴. In comparison, Crestani et al. found higher levels of secondary serum bile acids in children with FA compared to those with asthma.⁷ In our study, we identified that taurocholate (primary bile acid metabolism) and taurohyocholate (secondary bile acid metabolism) levels in children at < 1 year of life was significantly associated with increased risk of overall FA; and three metabolites in secondary bile acid metabolism (taurohyocholate, tauroursodeoxycholic acid sulfate and taurochenodeoxycholic acid 3-sulfate) were associated with an increased risk of cow’s milk allergy, which were consistent with the findings of Crestani et al. Secondary bile acids are derived from primary bile acids by the action of bacterial metabolism, which, in turn, may modulate the composition of the gut microbiome, leading to the development of allergy. It is also likely that bacterially modified bile acids may modulate T-cell differentiation and the generation of T regulatory cells, and thereby contribute to the development of allergy.

One novel finding of this study is that TMAVA is significantly and inversely associated with risk of overall FA, as well as with cow’s milk allergy and peanut allergy in specific; while the direction of association between TMAVA and egg allergy is consistent, it is not statistically significant after adjusting for multiple testing. TMAVA is involved in lysine metabolism. The study by Creastani et al. provided evidence that metabolites in lysine metabolism were different in those with FA vs non-atopic controls ⁷, although the individual metabolites identified in this previous study (lysine and N2-acetyllysine) were not the same as those we identified in the current study. TMAVA is derived from trimethyllysine through the gut microbiota and may inhibit carnitine biosynthesis and uptake, leading to free fatty acid oxidation inhibition and myocardial lipid accumulation and toxicity.³⁷ TMAVA-treated mice showed significantly altered lipid profiles compared with controls ³⁷. We observed that, in this study, TMAVA was positively correlated with galactonate, SMs (d17:1/16:0, d18:1/15:0, d16:1/17:0), and a ceramid (N-palmitoyl-heptadecasphingosine). Taken together, it is likely that TMAVA may have an impact on FA development by altering sphingolipid levels; however, further research is needed to confirm these findings.

Another novel finding of this study is that orotidine and 4-cholesten-3-one represent the top two metabolites that are associated with asthma but not with FA. Orotidine, derived from the dephosphorylation of orotidine-5-phosphate, is an intermediate in the synthesis of pyrimidine nucleotides. Orotidine was reported as a biomarker for increased risk of cardiovascular diseases in type 2 diabetes ³⁸. The biological mechanisms linking orotidine to asthma risk are unclear. One possibility is that increased orotidine is a marker of other metabolites and that the orotidine - asthma association as observed in this study may be due to its correlation with other metabolites. As an example, we observed that orotidine was significantly correlated with several metabolites that have been linked to asthma, including kynurenine (tryptopan metabolism, r=0.45)³⁹ and dimethylarginine (r=0.44).⁴⁰ Kynurenine itself was positively associated with risk of asthma at P < 0.05 in this study. 4-cholesten-3-one, or cholestenone, is a sterol which may play a crucial role in immune responses and human asthma.⁴¹ Cholestenone in cord blood can predict rapid growth and overweight/obesity⁴², which are risk factors of asthma development in childhood. Because cholestenone is produced by bacterial catabolism of cholesterol in the intestinal tract, it may be serving as a proxy indicator of the relative abundance of various microbiota.

When the combined allergic phenotypes were analyzed, we observed that children with both FA and asthma had altered androgenic steroids and lipids, which align well with the associations between these metabolites and risk of FA. Besides, two progestin steroids, which were significantly associated with time-to-asthma events (Table 3) and risk of FA (with the stringent definition, Table 2), separately, were marginally lower in children with both FA and asthma than those with neither phenotypes (P=0.004–0.006). A previous study reported that substantial reduction in 17 steroid metabolites were significantly associated with prevalent asthma, which was not fully explained by inhaled corticosteroid use.³¹ Finding from us and from this previous study suggest that the reduced steroid levels may be a fundamental characteristic of pathophysiology of asthma and FA. Further studies are needed to explore whether these metabolites are associated with other allergic phenotypes and are involved in the shared pathways leading to the “allergic march”. In comparison, we did observe that CMPF and hydroxy-CMPF, metabolites of furan fatty acid and markers of fish oil intake, were associated with reduced risk of having FA only, but they had no associations with risk of asthma. In comparison, Gurdeniz et al. ever reported that cord CMPF was associated with reduced risk of asthma but not with allergic sensitization ⁴³. Possible explanations for the inconsistency across studies include different time points when CMPF was measured and/or different genetic backgrounds of the studied populations, as a previous study demonstrated that fish oil intake and risk of incident asthma was modified by a fatty acid desaturase genetic polymorphism⁴⁴.

The strengths of this study include investigation of a multi-ethnic high-risk urban birth cohort (59% Black, 22% Hispanic) with metabolomic profiling performed during early childhood using state-of-the-art technology. All asthmatic cases were diagnosed by a physician at five years of age or older, ensuring temporal relationships. However, we could not exclude the possibility that some FA cases had developed allergy before metabolomic profiling. We addressed this limitation by removing those children with FA diagnosed before the time of metabolomic profiling, and our results remained largely consistent. We also acknowledged the following limitations of the study: 1) The number of FA cases, especially the food-specific allergic cases, was relatively small, which may have limited our power to perform subgroup analyses; 2) Aero-allergen IgE measurement was available only for a small proportion of children in this study, and thus we could not classify asthma into atopic vs non-atopic subtypes; 3) Although we adjusted for multiple potential confounders in investigating metabolomic associations with allergic phenotypes, residual confounding by unmeasured confounders is still possible; 4) A standard questionnaire interview by trained research staff was used to obtain data on clinical symptoms upon food ingestion and to collect covariates such as maternal smoking during pregnancy, maternal diet during pregnancy, breastfeeding and timing of first solid food introduction. Due to its retrospective nature, recall bias cannot be ruled out; 5) Untargeted metabolomic profiling in this study provides information on relative differences in metabolites across different allergic groups. In future, targeted profiling is needed to confirm quantitative differences in absolute levels of metabolites across different groups.

In summary, our study confirmed previous reports of reverse associations between lipid metabolites and risk of FA, and further suggested that the lipid - FA associations may vary by the specific food allergens triggering allergy. We also identified new biomarkers for FA (steroids and TMAVA) and asthma (orotidine and 4-cholesten-3-one). These findings, if further confirmed by future studies, have the potential to inform the development of novel preventive and treatment strategies for asthma and/or food allergies.

Key Messages.

• Androgenic, pregnenolone, and progestin steroid metabolites and lipid metabolites are significantly associated with FA, and the associations may vary by the types of FA.

• Orotidine and 4-cholesten-3-one represent the top two metabolites associated with risk of asthma, neither of which has an association with FA.

• Children with both FA and asthma exhibited an altered metabolomic profile during early childhood that aligned with that of FA.

Abbreviations

AD

Atopic dermatitis

BBC

Boston Birth Cohort

BMI

Body mass index

CI

Confidence interval

CMPF

3-carboxy-4-methyl-5-propyl-2-furanpropanoate

DAG

Diacylglycerol

DHEA-S

Dehydroepiandrosterone sulfate

EMR

Electronic medical record

FA

Food allergy

FDR

False discovery rate

FS

Food sensitization

Fs-IgE

food-specific immunoglobulin E

GEE

Generalized estimating equation

ICD

International Classification of Diseases

iNKT

Invariant natural killer T

OR

Odds ratio

PM_2.5

particulate matter less than 2.5 micrometers in diameter

PPV

Positive predictive value

PUFA

Polyunsaturated fatty acid

QC

Quality control

RSD

Relative standard deviation

SD

Standard deviation

SM

sphingomyelin

TMAVA

N,N,N-trimethyl-5-aminovalerate

1-Oleoyl-2-arachidonoyl-GPI

1-Oleoyl-2-arachidonoyl-sn-glycero-3-phosphoinositol